HVWXGLRV VREUH OD HFRQRPLD HVSDÔROD

-RE ORVV DQG XQHPSOR\PHQW GXUDWLRQ IRU ROGHU ZRUNHUV LQ 6SDLQ

$OPXGHQD %ODQFR 6DUD GH OD 5LFD

(((

http://www.fedea.es/hojas/publicado.html

Almudena Blanco and Sara de la Rica¤ Universidad del País Vasco November 4, 1999

Abstract

¤For contact: Sara de la Rica - Universidad del País Vasco. Avenida Lehendakari Aguirre, 83 - 48015 Bilbao (Spain) - Tfno: 94-6013783 - e-mail: jeprigos@bs.ehu.es

Abstract

The aim of this study is to analyze how unemployment a¤ects older workers, which is a demographic group of increasing importance in the Spanish population structure. The speci…c empirical issues we try to answer are (i) which are the main determinants of job loss for these workers, relatively to prime age workers, and (ii) which are the main determinants for these job losers to …nd a job again. We use a longitudinal sample of the Spanish Labour Force Survey (1992:I - 1997:2) and the …ndings suggest that whereas individual characteristics, such as quali…cation are important determinants for prime age workers in order not to su¤er a job loss, the impact of these characteristics for older workers is very small, and dominated by other issues, such as …ring costs or demand conditions. With respect to unemployment duration, we …nd that technical education helps prime age workers to …nd a job. However, for older workers education seems to have a negative impact in order to leave unemployment. (JEL J14, J64).

Keywords: elderly, job search, hazard.

1 Introduction

The phenomenon of population aging in most developed countries is not new. Since the early decades of this century, we are facing with an increasing percentage of older individuals (45 or more) relative to younger ones (14 years or less). Spain is not unaware of this phenomenon. Figure 1 shows this aging process very clearly. However, what it is new and in some sense alarming in the last two decades, not only for Spain but for most OECD countries, is the high pace that this process is taking. Figure 1 shows how the aging process is being accelerated since the early eighties. Furthermore, it is expected that whereas the ratio of the older population (65 and above) to that of working age in the OECD areas has already risen from 15 percent in 1960 to 21 percent in 1998, it will rise up to 35 percent by 2030 (OECD, 1998).

[Insert …gure 1]

This phenomenon implies a formidable change in the composition of the labour market, with fewer young people entering the labour market and an increasing proportion of older workers in pre and post- retiring ages. As the Secretary-General of the OECD, Donald Johnston, recently addressed,

”... meeting the challenge of ageing populations will require comprehensive reforms that address the …scal, …nancial and labour market implications for pensions, social bene…ts, and systems of health and long-term care. The goal must be to harness the skills and experiences of older people and to ensure adequate living standards for them without placing an unfair burden on younger people”. (OECD, 1998, page. 3).

Policy action must combine …nancial, economic and social measures to cope with this phenomenon. Given that older workers will become a greater proportion of the labour force in the following decades, one of the aspects that experts stress is the need for these workers to work for longer and hence reverse the recent early retirement patterns that many developed countries have experienced. The possibility for older workers to work for longer, however, requires a variety of reforms that ensure that more job opportunities are available for them and that they are equipped with the necessary skills to take them.

The employment problems that older workers face are, therefore, at the root of the whole analysis of the aging phenomenon and most European countries are developing active and passive measures to cope with them. In Spain, an example of active measures for workers older than 45 can be found in a new permanent contract for these workers, with tax reductions for employers (law 64/1997). Passive measures like assistance unemployment bene…ts for workers over 45 with family dependents also exist in our country. Someone might argue that the labour market situation of this group of workers is not that bad, given that the percentage of older workers that su¤er unemployment is lower than for other groups of population. Taking Spain as an example, the male unemployment rate for workers over 45 is about one third of that for workers under 24. However, this fact should not understate another one, that is that older job losers often …nd more serious di¢culties than younger workers to …nd another job. Figure 2 shows the duration of unemployment for di¤erent age groups and it can be seen that almost 40 percent of workers between 45-54 stay unemployed for more than two years, that this percentage rises to almost 50 percent when we look at workers over 54, whereas average unemployment duration diminishes greatly as we look at younger individuals.

[Insert …gure 2]

The di¢culty for older job losers to …nd another one is hence obvious, so policy action must therefore be directed towards those older workers that face the risk of unemployment and to those that are already unemployed. In order to dictate policy measures directed to this group of individuals, it is …rst necessary to know who among older workers face the risk of unemployment and which are the determinants for them to lose their job. Second, for job losers, we must know which factors contribute for them to remain unemployed. This empirical analysis will give us the clue to design policy measures that help older workers who face employment problems. This is going to be the main aim of our paper.

The incidence and duration of unemployment has already been analyzed for the Spanish economy by some authors. In particular, Bover et al (1997) concentrate on the e¤ects of unemployment bene…t and business cycle on unemployment duration for males younger than 64 years old. García (1997) analyses also unemployment duration for workersyounger than 55and for theperiod 1978-1993. Another example is Alba (1998) who concentrates on unemployment duration of young workers. Our study, therefore, must be seen as a complement to these ones, given that it analyses employment di¢culties of a speci…c group of population with an increasing importance in the demographic structure of our country.

The paper is organized as follows: Section 2 describes the data and presents a descriptive analysis of the sample we use for estimation. In section 3, we present the empirical analysis of which factors contribute that older workers lose their job. This analysis will be done not only for older workers, but also for a reference group of adult but younger workers (35-44 years) to capture which factors are speci…c determinants for older workers to be at risk of unemployment. Section 4 presents the analysis of unemployment duration for older workers. Finally, the last section is devoted to outline the most important conclusions derived from the empirical work, as well as possible indications for economic policy that in view of our results might help to shorten the duration of unemployment for older workers.

2 The Data

We use a longitudinal sample taken from the Longitudinal Spanish Current Population Survey (EPA enlazada). The National Bureau of Statistics interviews around 60.000 households quarterly and the process is repeated for the same individuals for six consecutive quarters. It is possible therefore to follow each individual for around one year and a half (six observations per individual). The period we analyze is 1992:1 - 1997:2. We get extensive information about each individual´s labour market situation, as well some information about the rest of household members.

As we said in the introduction, we are interested in the analysis of employment problems that older workers face, and in particular, in the determinants for older workers to lose their job and for these job losers to remain unemployed or rather to …nd another job. Given the low participation rate of older females in our country (less than 40 percent in 1997), we are left with a very small sample of older females who lose their job and either remain unemployed or …nd another job. In addition to this, the Spanish family structure where the male is in most cases the head and main earner of the family, makes that the labour market situation of older females are very much dependent of their husbands ´, and therefore, the labour market problems of older females are probably more adequately analyzed in a family framework context, which is out of the scope of this paper. Therefore, we restrict our empirical analysis to males.

However, we must take into account that during the recent past, voluntary retirement has been a quite standard labour market measure for workers over 55 in our country (see …gure 3). For workers that face the possibility of voluntary retirement, unemployment is in many cases a transitory agreement between employer and employee before voluntary retirement.

[Insert …gure 3]

Our aim in this study is to analyze job loss and unemployment duration of older workers that do not face the possibility of voluntary retirement, given that it is for these workers that unemployment is really a distressful situation. In …gure 4 it can be observed that it is not before 55 years old that Spanish workers start doing the transition unemployment - retirement. Besides this, the empirical analysis of the labour market problems of workers older than 55 is more complex, since as OECD (1998, page.. 331) notes, (i) it is not easy to distinguish between supply and demand factors, and (ii) measurement error problems are more likely to appear given that many of them classify themselves as unemployed whereas they are voluntary retirees, and viceverse. Finally, given that our methodology consists of comparing the labour market situation of older workers relative to that of adult younger workers, and that the latter do not face any retirement prospect, we have considered more appropriate to restrict our sample ofolder workersto those that are not a¤ected by the possibility of voluntary retirement. We have therefore delimited the age for older workers to be less than 56. In doing so, we avoid unobserved heterogeneity derived from di¤erent future prospects that older workers may face.

[Insert …gure 4]

With respect to the reference group (henceforth prime age workers) we have used workers that are between 35 and 44 years to capture the labour market situation of workers that are not a¤ected neither by problems derived from labour market entry, nor by older workers problems. As Toharia (1997, pp. 46) in his study of the Spanish labour market notes,”... the phase of integration into the labour market takes place until the age of 30-34. The central ages 35-44 represent both the ending point of the integration phase and the starting point of the exit place, and can be taken as reference points for both of them”.

In order to see a …rst approximation of the labour market transitions that both older workers and prime age workers experience, we can look at table 1. It describes transitions between the di¤erent possible states in the labour market (employment, unemployment and out of the labour force) in the interval of one year. We want to highlight that (i) the probability of job loss is higher among the group of prime age workers (4.40 %) than among older workers (3.47 %), (ii) older unemployed workers face more di¢culties to …nd another employment (25.8 % of them …nd employment in the interval of 1 year) than prime age workers (32.9 % of them …nd it in the same interval of time), and (iii) transitions to inactivity, both from employment and from unemployment, are more frequent for the group of older workers than for the reference group.

[Insert table 1]

Although our initial sample contains, as table 1 shows, 32785 individuals aged between 45-55 years and 34394 individuals in the reference group, in order to analyze job loss and unemployment duration we need to observe in the …rst place individuals that are employed in the …rst interview and either remain employed or lose their job. This restriction reduces our total sample in 12065 individuals. In the second place, for those who lose their job, we need to observe either complete or incomplete unemployment durations. After dropping those who experience directly the transition employment-inactivity (1373, which represents 2% of the sample), those individuals for whom we have missing observations before experiencing any transition and those who answer inconsistently to some of the questions, we are left out with our …nal sample, which accounts for 19421 older workers and 19877 prime age workers.

3 Empirical analysis of job loss

In this section we analyze which are the characteristics of older workers who lose their job relative to those of prime age workers, as well as the determinants for job loss.

In table 2 we present a preliminary description of the sample. As we are interested in looking at individuals a¤ected by job loss, we distinguish between those individuals that do not lose their job (and remain in it) along the period we observe them from those that although being employed at the time of the …rst interview, lose their job at some point while observed. From table 2 it can be inferred that for both age groups, workers who su¤er job loss are individuals with low education, low quali…ed and unstable (temporal) jobs.

[Insert table 2]

However, there are some di¤erences between job losers of both age groups that are noteworthy. In the …rst place, the increasing educational level attained in Spain in the recent past can be re‡ected in the percentage of job losers with no studies. For the older group, 27.75 % ofworkers have no studies, compared to 12.95 % for the prime age group. In the second place, the percentage of head of household among older job losers (91.25 %) is much higher than the percentage of heads among prime age job losers (78.65 %). This is an important aspect from the social point of view because given the low participation rate of older females, it is reasonable to expect that older males were the only labour income earners in the family, and hence job loss is likely to represent a very important economic distress not only for them, but for the whole household.

In order to know the determinants to lose a job for the group of older workers relative to prime age workers, we have estimated a logit model where the dependent variable takes the value of one if the worker loses his job at some point along the time they are observed and zero otherwise. Results are presented in table 3.

As independent variables we have included personal characteristics (age, education), family structure (marital status, head of household or not), measures of family income (approximated by the number of no economic dependents in the household), previous job characteristics (type of job (public/private/self employed,...), type ofcontract (temporal/permanent), hours ofwork (full-time/part time), tenure in the previous job, occupation) and some demand variables such as activity, male unemployment rate by province and the growth rate of the Gross Domestic Product¹.

[Insert table 3]

The …rst result that we would like to highlight is the di¤erent role that quali…cation, described in terms of occupation, plays in order to lose a job. For prime age workers, increases in quali…cation clearly decreases the probability of a job loss. This impact is however very small for the group of older workers. Only highly quali…ed white-collar workers (professionals) show a lower probability to lose the job with respect to non-quali…ed manual jobs. Among the rest of occupations, there seems to be no di¤erence in the probability to lose the job. This indicates that personal characteristics play a smaller role for a job loss among the group of older males than for younger adults. In the same line we can interpret the result we …nd with respect to education. For the older group, only very educated (university) or very low educated (no studies) show di¤erences in the probability to lose the job with respect to individuals with primary studies. For the prime age group, however, every increase in educational attainment diminishes the probability to lose the job. Again, personal characteristics seem to have a greater e¤ect for the probability of a job loss for younger workers than for older workers.

Other results re‡ect that most individuals that su¤er a job loss were working in a temporary basis, which seems very reasonable in our country given the high costs associated to laying o¤ a permanent worker. Besides this, it is likely that temporary workers ´jobs are more associated with discontinuous and/or seasonal activities. Among permanent workers, however, we can see the e¤ect that tenure on the …rm exerts for the probability of a job loss. Having a higher tenure seem to o¤er more protection to older workers with permanent contracts against job loss than to comparable prime age workers. In order to interpret this …nding, we must associate tenure with …ring costs, given that severance payments in Spain are proportional to the number of years that workers have been employed on the …rm². This result is thus telling us that …ring costs prevent against job loss especially to older workers. For prime age workers, the e¤ect of …ring costs for the probability of a job loss is smaller, perhaps because there are other personal characteristics, like quali…cation that also prevent against job loss.

Finally, demand conditions also a¤ect the probability of a job loss. Favorable economic conditions, translated in higher GDP growth or a low male unemployment rate diminishes the probability of a job loss, for both older and prime age workers.

In summary, we might say that for older workers the probability of a job loss does not seem to be a¤ected by personal characteristics, such as quali…cation but rather by other determinants, such as …ring costs or demand conditions. This seems to be a particular result for this group of age, given that when considering the determinants of a job loss for prime age adults, personal characteristics such as quali…cation and education help workers to retain the job. This might re‡ect a di¤erent behavior of employers with respect to layo¤s when treating with di¤erent age groups, in the sense that when deciding layo¤s younger workers are evaluated in terms of their personal characteristics, whereas older workers’ personal characteristics do not play any important role for determining who loses and who retains the job.

4 Duration of unemployment

4.1 Theoretical framework - The search model

The standard framework to explain unemployment duration is the search model presented by Mortensen (1970). This model can be used to explain unemployment duration for those who end up in another job.

It is based on the optimal job search strategy for an unemployed worker. It is assumed that the worker is actively looking for a job, but faces imperfect information concerning location and wages of available jobs. Therefore, in each period t individuals incur in search costs. Once a job o¤er arrives, each individual must decide whether to accept the or not. Therefore, the probability for an unemployed worker of leaving unemployment depends on the probability of receiving an o¤er ¸(t) and the probability of accepting it, which depends on the reservation wage, !¤. Denoting by the probability of leaving unemployment, we can de…ne it as

\[\mu (t) = _ {s} (t) [ 1; F (! ^ {\alpha}) ]\]

This is the structural expression of the hazard rate for unemployed workers. However, we will estimate the reduced form, where we consider variables a¤ecting both ¸(t) and !¤: We will not, therefore, be able to estimate separately neither which is the e¤ect of the reservation wage nor the e¤ect of ¸(t) on the hazard rate.

With respect to variables a¤ecting ¸(t) we must include those variables re‡ecting personal characteristics that make them attractive to employers, such as education, experience, as well as variables re‡ecting the demand side. With respect to variables a¤ecting !¤, we must include search costs and unemployment income.

4.2 Econometric model

Our sample of unemployed workers consists of workers that were employed in the …rst interview and lost their job at some point while observed. This means that these workers are entrants into unemployment from a job. This avoids potential stock sample bias³. We will use the so-called single-risk models, that allow to empirically analyze the duration of processes when only one type of risk is considered. We will consider the hazard to employment. Unemployment duration (T ) will be de…ned in months, given that although the survey is conducted quarterly, it is possible to de…ne it in months using other available information from the survey⁴.

We de…ne by the length of unemployment period for each individual i. Given the discrete nature of our data, and that we assume a proportional hazard parameterization⁵, we de…ne the hazard rate as

\[\begin{array}{r c l} & & 2 \\ h _ {i} (t) & = & \operatorname * {P r} \left[ T _ {i} < t + 1 j T _ {i}, t \right] = 1 i \exp^ {4} i \int_ {t} ^ {\mathbb {Z} ^ {+ 1}} \mu_ {i} (s) d s ^ {5} \\ & & 2 \\ & = & 1 i \exp^ {4} i \int_ {t} ^ {\mathbb {Z} ^ {+ 1}} (s) \exp f x _ {i} (s) ^ {0 -} g ^ {5} \end{array}\tag{1}\]

We also assume to be constant for . This allows the hazard to be written in the following terms:

h_i(t) = 1 exp [exp x_i(t)0¯ + °(t) ] (2) where captures duration dependence non-parametrically. Given this hazard rate, the likelihood of individual i can be de…ned as

\[L _ {i} = 4 h _ {i} (d _ {i}) \underset {t = 1} {\overset {d _ {Y}} {\longrightarrow}} [ 1 i h _ {i} (t) ] ^ {5} \underset {t = 1} {\overset {2} {\longrightarrow}} [ 1 i h _ {i} (t) ] ^ {5}\tag{3}\]

where is the observed duration for the individual (complete or censored) and is an indicator variable which takes the value of one if the observation is complete and zero if it is censored.

However, there is an alternative way to de…ne this likelihood that makes computations much easier. Following Jenkins (1995), we de…ne a variable which takes the value of 0 for all spell quarters except the exit month, in which case takes the value of one. Note that for stayers, for all spell months. Using this indicator variable, the likelihood function of each individual i can be written as

\[L _ {i} = \underset {t = 1} {\mathbf {Y}} ^ {\prime \prime} \frac {h _ {i} (t)}{1 i h _ {i} (t)} \underset {t = 1} {\# y _ {i t}} \underset {t = 1} {\mathbf {Y}} [ 1 i h _ {i} (t) ]\tag{4}\]

In addition to this, we estimate this model adding a term to correct for the presence of unobserved heterogeneity. Not introducing this term would imply not correcting for unobserved individual di¤erences, which might lead to bias, not only on the baseline hazard but also on the parameters associated to the explanatory variables. We include this term, as it is standard, in a multiplicative way. Denoting by the heterogeneity term, the hazard rate for discrete data can be rewritten as

\[h _ {i} (t) = 1 \text { i } \exp f _ {i} \exp [ x _ {i} (t) ^ {0 -} + ^ {\circ} (t) + \log (" _ {i}) ] g\tag{5}\]

The log-likelihood function for all individuals in the sample, introducing the heterogeneity term is written as

\[\log L = \sum_ {i = 1} ^ {N} \log f ^ {Z} \exp^ {2} f _ {i} ^ {\prime \prime} \int_ {t = 1} ^ {d _ {X}} \exp f ^ {\circ} (t) + x _ {i} (t) ^ {0 -} g ^ {5} d g \left(" _ {i}\right) i\tag{6}\]

We assume that "_i follows a normalized gamma distribution⁶. Hence the loglikelihood can be de…ned by the expression

\[\begin{array}{r l r} \log L & = & \underset {i = 1} {\overset {\mathbf {X} ^ {d}} {\log f ^ {4}}} 1 + \underset {t = 1} {\overset {3 / 4 ^ {2}} {\exp f ^ {\circ} (t) + x _ {i} (t) ^ {0 - g ^ {5}}}} ^ {2} \\ & & \underset {i} {\overset {2} {\operatorname{ci} ^ {4}}} 1 + \underset {t = 1} {\overset {3 / 4 ^ {2}} {\exp f ^ {\circ} (t) + x _ {i} (t) ^ {0 - g ^ {5}}}} ^ {3} \end{array}\tag{7}\]

4.3 Empirical …ndings

As we said when we described the theoretical framework to be used in the empirical analysis of duration of unemployment that ends up in employment, we must take into account variables related with the probability of receiving a job o¤er and with individual´s reservation wage. Among variables that a¤ect the probability of receiving a job o¤er we include personal characteristics that re‡ect accumulated human capital, such as education, tenure in the last job as well as age (to proxy experience) as well as some characteristics related to the former job (type of contract, activity). Demand conditions also a¤ect the probability of receiving a job o¤er, so we also include variables that re‡ect the business cycle. With respect to the reservation wage, wemust includeincomevariables. Given that wedo not havedirect measures of family income, we include approximations to this variable (number of household members that are not economically dependent). Approximations to individual income of unemployed workers are measured by an indicator that takes the value of one if individuals receive unemployment bene…ts⁷.

Figures 5 and 6 present the univariate Kaplan-Meier estimations of the hazard rate. It can be seen that instantaneous hazard rates are very low for the …rst unemployment periods, they increase up to the …fth month approximately, and from then onwards, the hazard decreases continuously. This pattern is quite standard in many empirical studies of unemployment duration. The low hazard rate in the …rst months is usually associated with a low search intensity at the beginning of the unemployment period which is translated into a low job …nding rate. The decreasing hazard from the …fth month onwards is normally associated to human capital losses associated to workers that remain unemployed. Nevertheless, note that the hazard rate is always higher for prime age workers than for older workers (see …gure 7), which re‡ects that older workers have more di¢culties to …nd a job after a job loss.

[Insert …gures 5, 6 and 7]

Table 4a presents estimations of duration models for both groups of unemployed workers. We present the results from such estimations when unobserved heterogeneity is not taken into account (columns I and III) and when it is taken into consideration (columns II and IV). Our comments will refer to the latter case, given the signi…cance of the term that controls for heterogeneity⁸. In any case, results of both estimations are very similar in terms of signi…cance and sign of coe¢cients of the explanatory variables, except for the coe¢cients of the duration dependence⁹, whose values are presented in table 4b.

[Insert tables 4a and 4b]

Turning to table 4a, we can see that in the …nal estimation there are some interactions between age and other explanatory variables. We started introducing interactions among all explanatory variables and age to see if the e¤ects changed as age increased even inside of each group. The …nal estimation presents only those interactions that were signi…cant (at 90 % of signi…cance)¹⁰. With respect to the results, we want to highlight in the …rst place the …ndings that concern the e¤ect of education on the probability of …nding a job. We can see that for the group of older workers education a¤ects very little the probability of …nding a job, and for those educational levels where the e¤ect is signi…cantly di¤erent from zero (vocational II), the sign is negative. This result contrasts with the e¤ect that education exerts on the probability of …nding a job for prime age workers, where education (especially vocational education) has a strong positive impact on the probability of …nding a job. The little impact of education for the group of older workers can in principle be understood since investment in human capital was done many years ago and it is reasonable to think that there may be other variables that a¤ect more the probability of …nding a job. However, the fact that for those educational groups for which education is signi…cant (vocational II) the sign is negative whereas for prime age workers this particular educational level has a strong positive e¤ect to leave unemployment suggests that for jobs where a certain level of technical quali…cation is needed, employers prefer to employ younger workers. This interpretation is reinforced if we look at the e¤ect of this educational level (vocational II) when it is interacted with age for prime age workers. The older these workers are, the less positive the e¤ect of this educational level is in order to …nd a job. We may think of various reasons why employers do not want to employ older workers for jobs that require certain level of technical quali…cation. It is possible that they consider that their human capital is obsolete or that given that they must give workers some training they prefer to invest in speci…c human capital in younger workers in order to receive higher future returns of such investment. Another surprising e¤ect concerning education, which is common for both age groups, is that upper secondary and university studies (the latter is not strictly signi…cant) have a negative impact for the probability of…nding a job. A …rst possible explanation for this result is that higher educated workers may have a higher reservation wage than other workers and hence their probability of rejecting job o¤ers is higher than for the rest of workers. However, another possible explanation is that given the impressive increase in the educational level that Spain has gone through during the last decades, there is a shortage of demand for jobs that require upper secondary or university studies.

Other results we can observe from table 4a are the following: (i) Having worked as a self-employed a¤ects the probability of …nding a job negatively, both for older and for prime age workers. It is possible that self-employed workers have less information about labour market possibilities given that they have worked as independent workers. (ii) With respect to the e¤ect ofeither personal labour income or family income of unemployed workers, we can see in the …rst place that receiving unemployment bene…tsa¤ectsnegatively and very strongly the probability of…nding a job. This is a very standard result in the empirical analysis of unemployment duration for many countries and for every age group. Personal income increases the reservation wage and so the probability of rejecting job o¤ers increases.(iv) Finally, positive demand conditions, represented either by a high growth rate of GDP or a low unemployment rate increase the probability of unemployed workers of …nding a job.

5 Conclusions

The aim of this study has been to analyze how unemployment a¤ects the group of older workers in Spain. Older workers, although do not exhibit high unemployment rates, su¤er higher proportions of long-term unemployed than other age groups. We limit our analysis to workers under 56 (and over 44) because in Spain from 56 years onwards we start to observe transitions from unemployment to voluntary retirement. For these workers, unemployment represents a transitory stage from employment to retirement. Unemployment is a severe and distressful problem particularly for those that can not consider the option of voluntary retirement and thus we concentrate in such group of older workers. The study has been done using a comparative group of prime age workers (between 35 and 44). This allows us to better characterize the labour market problems that older workers face compared to younger workers. The data used is a longitudinal sample of the Spanish Labour Force Survey and the period covered is 1992:I - 1997:II. The two speci…c issues we have empirically analyzed are (i) which are the main determinants for older workers to su¤er a job loss compared to prime age workers, and (ii) the determinants for these job losers to …nd a job.

With respect to the determinants of job loss we have found that in general terms individuals with low education, employed in relatively unstable (temporal) jobs are those who are more at risk to su¤er a job loss. However, a noteworthy …nding of our analysis is that when comparing the determinants ofjob loss for older workers with those ofprime age workers, we …nd that for the latter personal characteristics such as quali…cation, measured in terms of education and type of occupation help individuals not to su¤er a job loss. However, for older workers, the impact of quali…cation on job loss is very small. For these, the main determinants to prevent a job loss are issues such as high …ring costs and favorable demand conditions. A possible interpretation of this …nding is that human capital of older workers is rather obsolete given that it was acquired long time ago, even for those that are relatively more educated and whose jobs are quali…ed. If that is the case, personal di¤erences in quali…cation do not make real di¤erences in productivity, and hence the probability of a job loss does not depend on them. However, this interpretation implies that older workers have not received enough on-the-job training, which is likely to be the main determinant for human capital not to be obsolete. It is possible that we are in the face of a market failure case, given that employers have no incentives to invest in speci…c human capital for older workers because of their relatively short future working years. In this sense, we consider that it is necessary to design policy measures that encourage employers to invest in on-the-job training for the group of older workers in order to avoid that their human capital depreciates.

Relatively to unemployment duration, an important …nding is that technical education (vocational I and vocational II) have a very strong e¤ect to help prime age workers to …nd a job, but a negative e¤ect for older workers. This suggests that employers are reluctant to employ older workers for jobs that require certain level of technical quali…cation. A possible explanation of this is, in line of what has been said above, that human capital of older workers is rather obsolete, especially when it is referred to technical human capital. Fiscal incentives to employers when contracting older unemployed workers might serve to partially solve this problem.

Another interesting result that we have found is that for both age groups upper secondary education has a negative e¤ect to leave unemployment. Given the impressive increase in educational attainment that Spain has been going through in the last decade we think that this negative e¤ect may be due to an excess of supply or equivalently a shortage of demand of jobs that require such level of education.

Finally, the last result that we want to mention is that for both age groups workers that lose a self-employed job experience a greater di¢culty to …nd another job. This may be the result of their lower information about labour market possibilities given the type of job they used to have. Some policy measure might be taken in order to help these workers to leave unemployment, given that in many cases self-employed jobs are family business, and a job loss may entail the loss of job for many members of the household.

Data appendix

De…nition of variables used in the empirical analysis:

- Duration of unemployment: In order to create the dependent variable, duration of unemployment, we have considered the answer to the question ”When did you start looking for a job?” at the moment we see they lose their job and enter unemployment. From such answer we can infer duration of unemployment (in months) for entrants. Then we add three months for each interviewthat individuals report to continue unemployed. If they do not leave unemployment, they will be considered censored at the time we stop observing them. Complete durations for individuals whose unemployment period end up in employment are constructed by looking at the answer to the question ”When did you start working in your current job?” at the time we observe they start working.

- Education: Primary education corresponds to 5 years of education, lower secondary to 8 years, upper secondary to 11 or 12 years, vocational I refers to two years of vocational studies after primary education, vocational II refers to four or …ve years of vocational studies after primary education. Finally, university studies can entail 3 or more years of university studies.

- Number of no dependents: It is calculated as the di¤erence, relatively to household size, between the total number of household member and the number of economic dependent members (members of less than 16 years old or older members that are either out of the labour force or unemployed without receiving bene…ts).

- Unemployment rate: Male unemployment rate by province. The source is the Spanish Statistics Bureau (INE). For the analysis of job loss, we have used the unemployment rate of the moment of job loss. For the analysis of unemployment duration, this variable changes with time, but given that the information reported is quarterly, we have imputed the information given in the nearest publication to

that particular month.

- GDP rate of growth: We have used the annual rate of growth of GDP at constant prices that National Accounts publish quarterly. For the analysis of job loss, we have used the GDP rate of growth of the moment of job loss. For the analysis of unemployment duration, this variable changes with time, but given that the information reported is quarterly, we have have imputed the information given in the nearest publication to that particular month.

-Unemployment bene…ts: It is a dummy variable that takes the value of 1 if the individual receives UB and zero otherwise. It changes with time. The information that we have about UB is quarterly, so in order to convert it to monthly data we have imputed the information given in the nearest interview to that particular month.

Notes

¹See the data appendix for a full description of all variables. ²The exact number of pay days depends on the type of dismissal (collective or individual and fair or unfair). For more details, see Bentolila and Dolado (1994). ³The maximum unemployment period that any individual can be observed is 16 months. This prevents somehow the analysis of long-term unemployment for this age group. However, the objective of this paper is not to deep into the determinants of long-term unemployment for older workers, but to study which are the speci…c characteristics that help and prevent unemployed older workers to have a quick exit rate to employment relatively to adult but younger workers. ⁴See the data appendix for a detailed description of the construction of the variable duration of unemployment. ⁵The proportional hazard is a very common parameterization due to its advantages. On the one hand, it does not impose any restriction on ¯ whereas it guarantees the non-negativity of the hazard rate. On the other hand, the estimation and inference of these models is rather direct (see Kiefer (1988)). ⁶Ridder and Verbakel (1983) show that this speci…cation is less restrictive the more ‡exible the duration dependence is. They also show that the other standard de…nition of unobserved heterogeneity (discrete non parametric estimation), although in terms of goodness of …t it gives better results, its e¤ect over …nal results is very small. Given the computation complexity of such assumption, we have assumed the gamma-distribution. ⁷The variable unemployment bene…ts (UB) is a variable that changes with time. This allows us to capture the e¤ect of exhausting UB for the hazard. To see precisely how we have constructed it, see the data appendix. ⁸At the end of table 4a we have included the signi…cance of the heterogeneity variance for those models where heterogeneity iscontrolled for, aswell as the statistic which corresponds to the test of the likelihood ratio of both models (including the term of heterogeneity against not including it). This value must be compared to a (3.84 at 5%). This test is often used in this context (see Lancaster (1990)), although the second model is not, strictly speaking, nested in the …rst one.

⁹Not correcting for unobserved heterogeneity induces to underestimate the duration dependence and diminishesthe e¤ect ofexplanatory variableson the hazard rate (see Lancaster (1990)).

¹⁰Similarly, we introduced interactions between the log of duration and some explanatory variables but they were not signi…cant, so we decided not to include them in the …nal estimation.

REFERENCES

1. -Alba, A. (1998): ”Re-Employment Probabilities of Young Workers in Spain”, Investigaciones Económicas, 22(2), pp.201-224.
2. -Bentolila, S. and Dolado, J.J. (1994): ”Labour Flexibility and Wages: Lessons from Spain”, Economic Policy: A European Forum; 9(18), April 1994, pp.53-99.
3. -Bover, O., Arellano, M. and Bentolila, S. (1996): ”Unemployment Duration, Bene…t Duration and the Business Cycle”, Estudios Económicos del Banco de España, Departamento de Inestigación , n^o 57.
4. -García, J.I. (1997): ”Las Tasas de Salida del Empleo y el Desempleo en España (1978-1993)”, Investigaciones Económicas, 21(1), pp.29-53.
5. -Jenkins, S. (1995): ”Easy Estimation Methods for Discrete Time Duration Models”, Oxford Bulletin of Economics and Statistics, 57.1, pp.120-138.
6. -Kiefer, N.M. (1988): ”Economic Duration Data and Hazard Functions”, Journal of Economic Literature, Vol.XXVI, pp.646-679.
7. -Lancaster, T. (1990): ”The Econometric Analysis of Transition Data”, Cambridge University Press, Cambridge.
8. -Mortensen, D.T. (1970): ”A theory of wage and employment dynamics”, in E.S. Phelps et al., eds, Microeconomic economic foundations of employment and in‡ation theory. New York: W.W. Norton.
9. - OECD (1998), Perspectivas del empleo, 1998, OCDE, Ministerio de Trabajo y Asuntos Sociales. Subdirección General de Publicaciones.
10. -Ridder, G. and Verbakel, W. (1983): ”On the Estimation of the Proportional Hazard Model in the Presence of Heterogeneity”, A&E Report 22/83, Faculty of Actuarial Science and Econometrics, University of Amsterdam.

- Toharia, L. (1997), ”Labour Market Studies: Spain”, O¢cial publication of the European Community, serie n^o 1.

Table 3. Logit estimation for the transition Employment-Unemployment¹ Dependent variable: Job loss=1; No job loss=0

¹Absolute t-value in brackets. ²Dummy variable for this tenure has not been included in the prime age group estimation because there is nobody younger than 45 with a tenure upper than 30 years.

³ The constant term represents the reference individual, it is, a male with primary studies, married, whose last job was in the private sector with temporal contract, in a non skilled manual occupation and in a manufacturing activity.

Figure 1. Population distribution by age groups (1900- 1997). Figure 1: Source: Anuario Estadístico de España (1900-1991) and Encuesta de Población Activa (EPA) (1997). Age groups for 1997 are: 0-15 years, 16-44 years and more than 45 years.

Figure 2. Distribution of unemployment duration by age groups (1997:II). Figure 2: Source: Encuesta de Población Activa (EPA enlazada), 1992:I-1997:II.

Figure 3: Source: Encuesta de Población Activa (EPA enlazada), 1992:I-1997:II.

Figure 4: Source: Encuesta de Población Activa (EPA enlazada), 1992:I-1997:II.

Figure 5:

Figure 6:

Figure 7: